Properties from Multiple Regression using Matrices
Proof: Define the n × 1 vector E such that E = Y − XB. The goal of the least-squares method is to find the (k+1) × 1 vector such that ETE is minimized. Now note that
This last equality is true since BTXTY is a scalar and so
Setting the first derivative of ETE equal to zero, we get
This is essentially another proof of Theorem 1 of Least Squares for Multiple Regression.
Property 2:
Proof: This follows from Property 1 and Definition 1 of Multiple Regression using Matrices since
Property A: The hat matrix H is symmetric and idempotent, i.e. HT = H and HH = H. The same is true for I − H.
Proof: First we observe that (XTX)-1 is symmetric since
Now
This shows symmetry. We now show idempotency.
Property 3: B is an unbiased estimator of β, i.e. E[B] = β
Proof: By Property 1
and so we have
since it is assumed that E[B] = β.
Property 4: The covariance matrix of B can be represented by
Proof: Based on the assumption that var(εi) = σ2 for all i and cov(εi, εi) = 0 for all i ≠ j, it follows that cov(ε) = E[εεT] = σ2I. As we have seen in the proof of Property 3
Thus by Property 3
Hence
Properties from Multiple Regression in Excel
Property B: For any n × n matrix A and n × 1 vectors Y and Z
Proof:
Thus
Since Tr(E[C]) = E[Tr(C)] for any square matrix C, Tr(CD) = Tr(DC) and b = Tr(b) for any scalar b, we have
Property C:
Proof: We first note that by Property 1
By Property 3 of Multiple Regression Analysis in Excel and the above equality
Property 4: MSRes is an unbiased estimator of σ2, the variance of the error terms εi
Proof: By Property 1 and C
By Property B
Since Y = Xβ + ε, it follows that E[Y] = E[Xβ] + E[ε] = Xβ, and so we have
Putting it all together we have
Hi Charles,
Your site provides someone with only limited understanding great insight into statistical methods. Many thanks.
Please could you explain how you differentiate the matrices in Property 1?
Alan,
The matrices are as in Multiple Regression using Matrices
Charles